Forum Discussion
Altera_Forum
Honored Contributor
7 years agoThanks for the great response,
I also have one more questions. Consider we have II > 1 in our for-loop in the single thread mode kernel. Now let's say there is still some parallelism opportunity in the code and I would like to move into NDRange mode implementation. As far as I know, in NDRange, multiple threads are being scheduled to the for-loop body instead of loop trips. I know that II in NDRange mode is not specified at compile time and will be determined at runtime. But is there any chance that our ND-Range mode can deliver better II, and as a result better performance?